RSX "taped out"

[expletive]

I understand the RSX has been 'taped out'. What exactly does that mean? What can still change in its specs from this point forward? Only clock speed?

Incidentally, does anyone know what the final specs are for the 360?

Cliff B from Epic made some comments on how great the hardware was but couldnt go into detail becuase he was "under NDA". Any rumblings that published specs on the hardware may have changed at all one way or the other?

I'm probably digging too deep but the 'under nda' comment gave me the impression there were some announced hardware updates, changes, features, that have been (will not be?) made public.

J

 

[Carl B]

Hey Expletive, do you know where you may have heard that RSX taped out? I haven't heard that yet, but am interested in any information to that effect.

As for the final specs, I don't know of any change that would have occured to the 360 specs - there was talk back in the day of how the R500 yields were so good that ATI was raising the possibility of a clock increase, but I feel we would have heard something about it. Ah well, if Microsoft did raise something, with a launch in two months we'll be sure to hear about it soon.

But it's this RSX tape-out information that I need a link or something on.

 

[expletive]

I know i read it so i'll try and track it down. Thanks for the reply.

Incidentally, when would the RSX need to be taped out in order to make a Spring 06 launch of PS3?

J

 

[Carl B]

I know i read it so i'll try and track it down. Thanks for the reply.

Incidentally, when would the RSX need to be taped out in order to make a Spring 06 launch of PS3?

J

When would it need to be taped out? Well... wouldn't need to be taped out yet. But the sooner the better if they want to have any of these things in quantity before launch. Also, with the R520 drama lately, you never know when you might need a re-spin of the design.

But if you could find anything indicative of tape-out, that would be great.

 

[expletive]

When would it need to be taped out? Well... wouldn't need to be taped out yet. But the sooner the better if they want to have any of these things in quantity before launch. Also, with the R520 drama lately, you never know when you might need a re-spin of the design.

But if you could find anything indicative of tape-out, that would be great.

Given that a google search didnt produce any results, i must have come across it from an unreliable source. News like this would have likley produce a whole list of sites reporting it. I'll keep looking through my history to see where i may have come across it though.

J

 

[hugo]

No signs of it on Nvidia's site as well.With such big news like this they would have probably made a press statement by now.

 

[jonnyp]

Given that a google search didnt produce any results, i must have come across it from an unreliable source. News like this would have likley produce a whole list of sites reporting it. I'll keep looking through my history to see where i may have come across it though.

J

Pretty silly making a thread like this with no source and then not even remembering where you read it and then stating that you must have gotten it from an unreliable source. Good going! hehe

 

[expletive]

If you read the original post the crux of my question was more along the lines of what does 'taped out' actually mean and what can change after ANY part has been 'taped out'. The status of RSX being actually 'taped out' is not intrinsic to what i asked.


J

 

[slider]

Thread title disagrees!

 

[Shifty Geezer]

'Taped out' means reading for fabricating, and nothing can change other than clockspeeds. You won't get more ALU's or features being added once the Die is ready for printing.

 

[BlueTsunami]

Well it seemed like the thread starter already assumed the RSX was taped out but was inquiring on what it means to be taped out. It doesn't seem like he was stating news on the RSX being taped out. Just a misunderstanding.

 

[slider]

Well it seemed like the thread starter already assumed the RSX was taped out but was inquiring on what it means to be taped out. It doesn't seem like he was stating news on the RSX being taped out. Just a misunderstanding.

True, I guess see Shifty's post above then.

I'm assuming such a thing as "small changes" to the GPU can't be done easily? Would a change in one part of a GPU architecture impact all over the chip?

Fundamentally the design of the GPU couldn't change, right? And if not, ignoring the clockspeed, what is the easiest thing to change or is that dependent on the initial architecture. As you can see I'm still getting used to GPU's.

 

[Carl B]

Ah well, so the conclusion is no tape-out as yet. Back to waiting mode then. I don't expect that we'll be getting much in the way of RSX specs until they tape it out, get some pre-production silicon, AND determine that their yields at whatever given spec will be acceptable.

Since the word is Sony plans to go all-out at the CES show in Vegas this coming January, that sounds like it lines up with the timeline on which those things would have taken place.

Maybe we'll get some architectural insights before that though...

I must say, the longer it takes to tape out, the more intriguing it becomes to me.

 

[Shifty Geezer]

'RSX's tape out' topic came up a while back. IIRC there was talk of a probable tape out in September from a technical source but I don't remember any official word on the matter.

 

[Carl B]

Well exactly. That's why I thought this topic when posted was right on schedule for the supposed September tape-out, but alas it was only a dream...

I think it was NVidia themselves that said it would be 'late summer.'

 

[Guden Oden]

I'm assuming such a thing as "small changes" to the GPU can't be done easily?

With the disclaimer that I don't pretend to be any kind of an expert on the subject, here goes:

As stated previously in the thread, after tape-out has occurred, nothing can change. Not without "taping out" again (ie: producing a new mask, or set of masks to be used for photoetching the silicon wafers that'll end up as hopefully fully functional GPUs). Producing these masks is mighty expensive (tens of thousands of buckaroos), so it's not anything any manufacturer wants to go through numerous times. Sometimes it is neccessary though; once first silicon is returned from the fabs, the chips have to be tested to see if they actually function as expected. Very minor errors can be fixed on a chip by zapping it with a kind of electron microscope/gun instrument (sort of like a microsoft windows hotfix for hardware ), but this isn't practical to do for more than a handful of chips; too time consuming.

Thus, minor revisions will then be performed to one or several of the masks and new ones will be produced, and then sent back to the fab to make a new batch of test chips. Everybody will then pray the process will not have to be repeated any more times.

Would a change in one part of a GPU architecture impact all over the chip?

Strictly speaking, no. Since at this stage you should have already designed all functionality in your chip, any changes will only correct errors in the already agreed upon design.

what is the easiest thing to change or is that dependent on the initial architecture.

That's way too complicated a question for me to answer. It will also probably vary from chip to chip, but I hear some metal layer features only require changes in one of the whole bunch of masks that are required to design an advanced IC. Thus, those kind of changes would be the easiest - not that this answer is likely to help you in any way! It's far too simplified and nebulous...

 

[overclocked]

Well exactly. That's why I thought this topic when posted was right on schedule for the supposed September tape-out, but alas it was only a dream...

I think it was NVidia themselves that said it would be 'late summer.'

I remember that also, August/September would seem likely IMO. If its avalability thats concerns thats something else but from a dev POV i dont think it would matter much, if anything.

 

[Carl B]

I remember that also, August/September would seem likely IMO. If its avalability thats concerns thats something else but from a dev POV i dont think it would matter much, if anything.

Yeah I agree, I don't think development of launch titles will be impacted much at all, if any. It's more just my own desire for a release of the architectural specifications.

 

[Simon F]

'Taped out' means reading for fabricating, and nothing can change other than clockspeeds. You won't get more ALU's or features being added once the Die is ready for printing.

Clock speed is something you have to design in as well so, apart from some minor variations, it's not going to change either.

 

[expletive]

When nvidia provides 'ultra' and 'gt' and '<no suffix>' parts for the same chip (for example the 6800 series), the differenent versions have varying clockspeeds and also varying numbers of pipes. My understanding is that all these chips come from the same batch. Is this correct? How do all these variations end up coming from the same batch of chips, particularly the varying number of avialable pipes?

J